Skill Claude CodeCodex
Creates specs before coding. Use when starting a new project, feature, or significant change and no specification exists yet. Use when requirements are unclear, ambiguous, or only exist as a vague idea.
Skill Claude CodeCodex
Use for structural code search with ast-grep when plain text search returns too much noise or when matching syntax shape matters.
Skill Claude CodeCodex
Drives development with tests. Use when implementing any logic, fixing any bug, or changing any behavior. Use when you need to prove that code works, when a bug report arrives, or when you're about to modify existing functionality.
Skill Claude CodeCodex
Handles failed tool calls safely. Use when a tool, command, MCP call, filesystem operation, network request, or agent action fails and you are considering a retry or recovery step.
Skill Claude CodeCodex
Stop agent work until a human gives a closed decision. Use when a task is blocked on human choice, approval, credentials, destructive action, ambiguous scope, or review/security findings. In OpenCode, prefer the native question tool; elsewhere emit WAITFORUSER: and stop.
Skill Claude CodeCodex
Cierre de turno ultracorto para el bridge dotmesh-watch. Termina el turno final con una línea WATCH para que el resumen quepa en el reloj Garmin. Explícita y dirigida por hook; no se auto-aplica.
Agent Claude Code
Implementation with full tool access. Use when there is an approved plan and the work needs to land in code. Follows incremental-implementation, tests after each slice, and invokes review/security at gate points.
Agent Claude Code
Reviews a markdown draft for markdown formatting, clarity, structure, and voice. Returns OK or a list of issues. Read-only — flags, does not rewrite. Use proactively after drafting each major section of prose and before finalizing a document.
Agent Claude Code
Verifies mathematical work with SymPy via Python one-liners. Returns formal verification result. Use proactively when the task involves mathematical derivations, formulas in decision models, or mechanism properties.
Agent Claude Code
Turns ideas into specs and implementation plans. Markdown only, no execution. Use proactively when starting a feature, when requirements are ambiguous, or when a change spans multiple files and no spec exists yet.
Agent Claude Code
Reviews a diff or code fragment with five-axis quality criteria. Returns OK or a list of issues with severity. Read-only — flags, does not fix. Use proactively immediately after writing or modifying code, before moving on.
Agent Claude Code
Reads a review thread and the surrounding document context, then proposes a reply or edit as a message.posted event with commit:null. Never edits the document body and never commits — the principal applies the proposal and creates the git commit. Use proactively when the principal fans out thread review tasks in…
Agent Claude Code
Security audit over diff, dependencies, or code fragment. Returns CLEAR or issues with CVSS and mitigation. Use proactively as a commit gate before committing security-sensitive changes (secrets, auth, input, deps, shell, network), not per-slice.
Command Claude Code
Initialize this workspace for the dotmesh agent system. Like /init but also wires shared skills and the AGENTS.md ↔ CLAUDE.md import.
Command Claude Code
Autonomously manage the Git lifecycle from sync and branch setup through atomic commits, push, and PR.
Hook Claude Code
Runs before the agent uses a tool for Bash, Write, Edit, MultiEdit, NotebookEdit, AskUserQuestion and Skill tool calls, executing block-dangerous-git.sh, pretooluse-approve.sh, remind-review-gate.sh, remind-load-skills.sh, pretooluse-ask.sh and handoff-signal.sh (7 commands). From PabloCoello/dotmesh.
Hook Claude Code
Runs when you submit a prompt, before the agent sees it, executing userpromptsubmit-watch.sh. From PabloCoello/dotmesh.
Hook Claude Code
Runs when the agent finishes a response, executing stop-reprompt.sh. From PabloCoello/dotmesh.
Hook Claude Code
Runs when the agent sends a notification for idle_prompt and permission_prompt tool calls, executing notification-notify.sh (2 commands). From PabloCoello/dotmesh.
Hook Claude Code
Runs on the StopFailure event, executing stopfailure-notify.sh. From PabloCoello/dotmesh.
Hook Claude Code
Runs when a session starts, executing herdr-agent-state.sh and remind-herdr-skill.sh (2 commands). From PabloCoello/dotmesh.
Settings file Claude Code
Agent settings declaring 6 hook events (PreToolUse, UserPromptSubmit, Stop, Notification).
Agent
You operate in the dotmesh engineering flow (AGENTS.md). Follow the skill flow on your own initiative and delegate to subagents proactively — stay a thin orchestrator while the workers carry the load in their own context.
Agent
Agent "scribe" from PabloCoello/dotmesh, covering scribe — dotmesh review persona, cycle, routing, response contract and invariant.